This is Scale & Strategy, the newsletter that paints you a picture of the future, before it happens
Here’s what we got for you today:
World models finally have a job, and it involves dodging elephants.
How to use seasonality adjustments in Google Ads to actually improve performance
World models finally have a job, and it involves dodging elephants.
As hype around world models keeps building, Waymo just gave the tech something concrete to do. On Friday, the robotaxi company unveiled the Waymo World Model, a generative system for large-scale, hyper-realistic autonomous driving simulations. It’s built on Google’s Genie 3, the latest version of its world model framework.
The architecture lets engineers modify scenes using plain language. Change the time of day. Add a storm. Generate an entirely new synthetic city. It can also turn real dashcam footage into simulated environments for training.
Waymo is basically taking Genie’s photorealistic, interactive 3D environments and applying them directly to driving. That means it can simulate rare edge cases like:
An elephant standing in the middle of the road
A tornado chasing a car down the freeway
A tropical city inexplicably covered in snow
The point is not cinematic flair. It’s stress-testing autonomy against extreme, low-probability events. If the system can handle the absurd, it should handle the mundane. In theory, this could help Waymo expand into new geographies and better prepare for natural disasters.
The timing is notable. The company has faced renewed scrutiny after one of its vehicles struck and injured a child in California. At a recent hearing, Chief Safety Officer Mauricio Peña said Waymo uses overseas remote operators for guidance in certain scenarios, though they don’t directly control the vehicles.
In its announcement, Waymo said simulating the “impossible” helps proactively prepare the Waymo Driver for rare and complex scenarios.
Zooming out, enthusiasm around world models is accelerating. Google recently introduced Project Genie, an experimental prototype built on Genie 3 that allows users to generate and explore interactive virtual worlds. Startups led by AI heavyweights like Fei-Fei Li and Yann LeCun are reportedly raising large rounds as investors look for the next frontier beyond LLMs.
For years, critics have argued that scaling language models alone won’t get us to AGI. Systems need an internal sense of physics and real-world dynamics. World models aim to fill that gap.
The existential question remains the same: is this just elegant theory, or does it actually matter?
Waymo might be offering one of the first real answers. Instead of debating AGI timelines, it’s using world models to train cars not to hit things. For once, the hype cycle has a steering wheel.
If your revenue is growing while your calendar stays packed, you haven’t hit a growth problem. You’ve hit a time constraint.
The average company loses more than 20% of productive capacity to meetings, approvals, and coordination drag, exactly where growing businesses leak profit and momentum.
The Executive’s Guide to Saving 10+ Hours Per Week reveals how operators reclaim time without adding headcount. These strategies will free you to focus on pricing, capital allocation, and strategic bets that unlock the next stage of growth.
Inside: • The hidden “calendar tax” slowing scale • Why decisions bottleneck behind founders • How operators build systems that move without them
And when you’re ready to turn insight into execution?
BELAY pairs you with a seasoned Executive Assistant who absorbs the operational load, helping leaders move from $5M → $15M without burning out or bloating overhead.
How to use seasonality adjustments in Google Ads to actually improve performance
One minute you’re drowning in holiday conversions. The next, it’s tumbleweeds and lonely clicks.
Google’s Smart Bidding is good. It’s not psychic. And that’s where seasonality adjustments come in.
What are seasonality adjustments?
They’re basically a manual signal you send to Google to warn it about short-term swings in conversion rates.
If you’re running something temporary, like a flash sale or promo, you can tell the system to expect a spike (or a drop). That way, strategies like Target ROAS don’t freak out and recalibrate around a short-lived surge.
AI is powerful. It still needs context.
Why seasonality adjustments matter
Smart Bidding leans heavily on historical data. The problem? When behavior shifts fast, history becomes useless.
Without adjustments:
Google might underbid during a massive surge and leave money on the table.
Or worse, it could overbid for weeks after your sale ends because it thinks the spike is the new normal.
Seasonality adjustments keep the algorithm from overreacting to temporary chaos.
When to use them
Think short-term and sharp. One to seven days is the sweet spot.
Examples:
Holiday promos
48-hour flash sales
Major product launches
They’re also useful in the opposite direction. If you know performance will dip, like the dead zone between Christmas and New Year, you can proactively scale expectations back.
Use them like a scalpel, not a sledgehammer.
When not to use them
Don’t use them for long, predictable trends like summer travel season or Q4 retail ramps.
Smart Bidding is built to handle gradual shifts on its own. If you interfere too often, you risk confusing the system and degrading performance.
Only step in when the change is sudden and temporary.
How to set them up
You’ll find seasonality adjustments in the “Tools” menu under “Budgets and bidding.”
Be specific with naming. Future-you will appreciate it.
You can apply adjustments to specific campaign types, which is ideal if only one product line is on sale while the rest of the account stays stable.
And treat every adjustment as a test. Once the event ends, review the data. Did the lift match your expectations? Did performance stabilize properly afterward?
Also, communicate with your team. If everyone understands why you’re adjusting the baseline, reporting gets a lot cleaner and you avoid those awkward “why did ROAS spike?” conversations in the monthly meeting.
Because the goal isn’t to outsmart the algorithm. It’s to keep it from outsmarting itself.
.png)
